0.2 0.4 0.6 0.8 1 1.2 1.4
0 0.1 0.2 0.3 0.4 0.5 0.6
PSfragreplaements
Overshoot
D
O S [% ]
Figure 5.22: Behavior of eah step response harateristi parameter with
re-spetto the damping ratio
Sinus response test
Aording tothistest,thebehaviorofanatuatorinthefrequenydomainan
beidentied. Theinputforsuhatestisasinusoidalsignalwithaninreasing
frequeny and adened amplitude.
TheBode diagramisthe mostimportantharateristiofthis test,gure5.23.
Thedamping ratioandthetime onstant arethesigniantparameters
hang-ing the bode diagram of the seond-order transfer funtion. The eet of the
damping ratio anbesummarizedasfollows: ifthedamping ratioof a
seond-order transfer funtion is less than 0.707, a resonane ours at the natural
frequeny of the system. The natural frequeny is
ω 0 = T 1
w
. Inversely, if the
damping ratio islargerthan0.707, no resonaneours. Theeet ofthetime
onstant (
T w
) is on the plae where the bode diagram falls o. By inreasingT w
,the falling happensinthe low frequenies.Extension of the methodology
Aording to theV-model, gure5.1, therequirementsof a large-sale system
an be broken down into the requirements formulated on the design variables
ofomponents(System Design). Withtheintrodution oftheveriation tests
in the last setion, the dependeny graph will be extended. In the extended
dependeny graph, a new level is added, the so-alled omponent design.
A-ordingly,the so-alledVeriationVariables level, whereveriationvariables
of eah veriation test are indiated, are dened in the same level of the
Subsystem level in the original V-model, gure 5.24 . Below this, the physial
Components Details areonsidered, whihshouldbespeied bythe
manufa-turer. In this way, we formulate the requirements on theveriation variables
from the requirements of the large-sale system. The way inwhih the
atua-tor isonstruted physiallywillbe surrendered tothe manufaturer. But the
0.3 0.5 1 2 3 4 -6
-5 -4 -3 -2 -1 0
0.3 0.5 1 2 3 4
-100 -50 0
PSfragreplaements
Frequeny [Hz℄
Frequeny [Hz℄
Magnitude[dB℄
Phasein[
◦
℄Figure 5.23: Bode diagram of a transfer funtion witha damping ratio larger
than 0.707
manufaturer annotdeliveranykindofphysialatuators, rather therealized
atuator shouldbeinthe form ofthe predeessors.
Figure5.24: Extended V-Modelfor thedesign ofatuators
Summary
Inthis hapter, we presentedthe robustdesign methodfordesigningatuators
and ontrol systems, whih an ontrol the omplexities in the development
proess and nd optimized intervals for the design variables,rather than only
one optimal set of the design variables. We have shown that the method is
based on the so-alled V-model in the system engineering. The method also
inludesthreeenablers,whiharedependeny graph,Bottom-Upmappingand
Top-Down mapping. Thedependenygraph manages theknow-howinthe
de-sign proedure andisagraphinwhihthe measurable system,subsystem,and
omponent properties are depited as verties. Their dependeniesor
intera-tions were shown withan edge, whose diretion shows whih subsystems' and
omponents' properties interat with eah other and inuene the measurable
system properties. The Bottom-Up mapping is thequantitative evaluation of
theinueneoftheinputsontheoutputs. Forthispurpose,weneededamodel
whih maps the inputs to the outputs. This model wasthe atuator and the
ontrol system model developed in the last two hapters. We explained the
neessity of lots of simulations on this model, whih provide information for
the Top-Down mapping. As a onsequene, we have laried the idea of the
Design of Experiments (DoEs) and how we an nd the best method ofDoEs
based on the disrepany fator. We also used the artiial neural networks
and support vetor mahines as meta-models to model our original atuator
and ontrol systemmodel. Thiswasdonetoaelerate theoptimization
proe-dure. Thisoptimizationproblemwasto ndthelargestbox, so-alledsolution
box, inside of the solution spae of the design variables. The solution spae
integrated dierent requirements fromdiverse disiplines. The edgesof the
so-lution boxservedasindependenttarget regionsforomponent properties. The
solution boxalso enhaned the probability of nding a valid design in the
so-lution spae. Welariedthatwehaveto onverttherequirementsformulated
on the design variables into the requirements whih should be formulated on
theveriation variables. Asaonsequene,weexplained dierent veriation
tests andvariables for thedesign variables ofan atuator.
6 Appliation
6.1 Results for robust designing of the rear steering
system
Inthis hapter, we applytheexplainedmethodintheprevioushapterfor
for-mulating requirementsonthedesignvariablesoftheontrolsystem,introdued
inhapter 4,andthe design variablesof ARSatuators, introdued inhapter
5. The rst step is to determine the dependeny graph. In this dependeny
graph, the relations between design variables of ARS atuator and theontrol
system and the objetive driving targets (CVs) are demonstrated. Moreover,
therelationsbetweenCVsandtheassessmentindies(AIs)aredepited,gure
6.1 .
Figure 6.1:Dependenygraphforthe ARSatuator,thelateraldynamis
on-trol system, andthe vehile genes
Asanbeseen,thedesignvariablesofeahfuntionoftheontrolsystemand
the ARS atuator and the genes of the vehile inuene the determined CVs.
For example, all design variables of thestati feedforward ontrol of ARS an
only eet the stationary objetive driving dynamis performane measures,
namely BRWCand QSSC.
Identiation of design variables
In the seond step, we have to identify the variables of the ontrol system
and the ARS atuator on whih the requirements have to be formulated. The
relevant parameters areasfollows:
Table6.1: Designvariablesof theARS atuator andits assoiatedontrol
sys-tem
System Parameter Desription
v trans v max i ARS,stat.,min
Stat. Feedforward
i ARS,stat.,max
Charateristi urve forthe stationary
feedforward (gure 3.2)
T f ω f
Dyn. Feedforward
D f
Fators forthedesired transferfuntion
G ψδ ˙ f ,des
(equation??)Feedbak
K p
Feedbakontroller proportionalgain (equation 3.45 )K p
D T w T d
Atuator LTItransferfuntion paramters
(equation 4.1 )
F max v noload v (F max )
Atuator fore-speed harateristi urve
(gure 6.2 ) Atuator
δ r,max
Maximum permitted rearsteering wheelanglem
Vehilemassrearload
Vehilerear loadratioh
VehileCG-heightGenes
J z
VehileInertia inz axisIt should be onsidered that the harateristi urve of theatuator has to
have aunitformlikethe one ingure6.2forthedesign regardingthe
ounter-Aordingly, we deal with three dierent groups of design variables. The
rst groupontains the genes of a vehile introdued in table 6.1 . Theyhave
their nominalvalueswhiharealreadypredened. A marginisleftfor themin
order toensuretherobustnessofthedesignedARSatuator anditsassoiated
ontrol system logi with respet to the hanges inthe vehile genes whih is
the ase, if we deal with dierent derivatives. The seond group inludes the
design variables oftheARS atuatorandthe thirdgroup ismadeofthedesign
variables ofthe ontrol systemlogi relevant totheARS.
Figure6.2: QSSC-CV
Forallthesedesignvariables,werstsamplethedesignspae. Thesampling
method is sramble Halton sequene in this appliation and the number of
sampling pointsis 4000.
Theonsidereddriving dynamis performane measures(CVs)arelisted
be-low:
Table6.2: Considered objetivedriving dynamis performane measures
Maneuver CV
EG H,nl
QSSC
a y,max
( ˙ ψ/δ H ) stat,70 ( ˙ ψ/δ H ) stat,190
WEAVE
( ˙ ψ/δ H ) stat,max β max
F z,min
SWD
SF M AX T eq, ψ/δ ˙
H
(H/H 0 ) ψ/δ ˙
H
CSST
T a y /δ H
BRWC
∆ ˙ ψ 1s
For eah ofthese CVs, the simulation of the whole systemis exeuted 4000
times. Then, individual ANNand SVM are trained byeah output (CV)and
a set of inputs (designvariables). Some results of ANN training areshown in
gure 6.3and gure6.4.
0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34
0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34
PSfrag replaements
WEAVE -
( ψ/δ ˙ H ) stat,max
Target
Neuralnetworkpredition
(a)
R 2 test
=0.985R training 2
=0.990-4 -3 -2 -1 0 1 2 3 4
-4 -3 -2 -1 0 1 2 3 4
PSfrag replaements
BRWC
-ψ ˙ ∆1s
Target
Neuralnetworkpredition
(b)
R 2 test
=0.920R 2 training
=0.949Figure6.3: Regression plots for CV
ψ ˙ ∆1s
and( ˙ ψ/δ H ) stat,max
8 8.5 9 9.5 10 8
8.5 9 9.5 10
PSfrag replaements
QSSC -
a y,max
Target
Neuralnetworkpredition
(a)
R 2 test
=0.940R training 2
=0.9560.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
PSfrag replaements
CSST -
T
eq, ψ/δ ˙ H
Target
Neuralnetworkpredition
(b)
R 2 test
=0.960R 2 training
=0.975Figure6.4: Regressionplots for CV
a y,max
andT eq, ψ/δ ˙ H
Asweanseefromgure6.3andgure6.4,themodelqualityofthetraining
data (
R training 2
) is larger than the the model quality of the test data (R 2 test
),whih means, there is no overtting. The model quality of all omputed
re-sponse surfaes for all CVs is more than 0.9, whih means, the requirement
identied inthe last hapterregarding the model quality ismet. The
mislas-siation of eahSVM for eahCV islisted below:
Table 6.3: Mislassiation ofall trained SVMfor eah CV
CV Constraints of CVs
EG H,nl
1.1a y,max
1.1( ˙ ψ/δ H ) stat,70
1.1( ˙ ψ/δ H ) stat,190
8( ˙ ψ/δ H ) stat,max
4.2β max
12.8F z,min
13.7T eq, ψ/δ ˙ H
15
(H/H 0 ) ψ/δ ˙
H
11.5
T a y /δ H
12∆ ˙ ψ 1s
12.1All the mislassiations are less than 15; hene the requirement regarding
themislassiation introdued inthe last hapter issatised.
Inthethirdstep,wehavetosetonstraintsontheCVs. Bysettinglowerand
upper bounds on eah CV, the solution spae will be formed. As mentioned
before, the largest volume of the solution box is unique. But there ould be
more than one box in the solution spae whih has the largest volume. For
instane, let uslook atthesolution spaeof twodesign variables,theatuator
transportdelay
T d
,andthenaturalfrequenyfatorofthedynamifeedforward ontrolw f
,gure6.5.Figure 6.5:Solution spae of the atuator's transport delay and natural
fre-queny fatorof thedynamifeedforward ontrol
Based on the introdued optimization algorithm inthe last hapter, the
al-gorithmseeksasolutionboxwiththemaximumsizeanddoesnot onsiderthe
meaning ofdesign variables. Forinstane, thesolutionboxillustrated ingure
6.6 restrits the lower bound of the transport delay interval. It means, the
atuator transport delay mustnot begin fromzero, whih implies slow
perfor-maneoftheatuatorregardingitsinput. Butontheotherhand,theexibility
for the parameterization ofthe natural frequenyfator hasbeen improved by
extending its permissible interval with respet to this solution box. However,
suh a solution box is undesirable, as we require a high-performane atuator
preferablywithnodelayandofoursemoreexibilityfordesigninganatuator.
Figure 6.6:Solutionboxwiththe largestintervalfor thenatural frequeny
fa-tor ofthe dynami feedforward ontrol
Balaning design variables of the rear steering atuator and the
ontrol system logi
Therefore,thealgorithmhastobeforedtoavoidsuhundesirableomputation
eets. As a result, we have to balane the atuator and the ontrol system
logi designvariables before starting theoptimization proedure. Aordingly,
we xthe lower bound of theatuator transportdelay
T d
to zeroinorder notto allow the optimization algorithm to nd a boxwhih restrits the atuator
performane, i.e. an interval for the transport delay is alulated whih does
not start fromzero. Basedon the samereason, we also xthelower boundof
thetimeonstant
T w
oftheatuator transferfuntion. Thelowerboundofthedamping ratio
D
hasalso to be xedto 0.707 inorder to prevent a resonanemagniation [46 ℄. In order to prevent limiting the maximal atuator power,
the upperboundsof parameters
v noload
andF max
havealsoto be xedto theirdened uppervalues.
Consequently, all onstraints demanded for nding the largest box an be
summarized mathematially asfollows:
Ω⊆Ω max ds
µ(Ω)
subjetto
f i ( x ) ≤ f c,i , i = 1, ..., n 1 g j ( x ) = 1, j = 1, ..., n 2 h k ( x ) ≤ h c,k , k = 1, ..., n 3
for all
x ⊆ Ω
(6.1)where
f i
isa predited funtion of thei
-th objetive driving dynamisper-formane measures by ANN,
g j
is a lassiation of thej
-th objetive drivingdynamis performane measures and
h k
represents additional boundaryon-straintsformulated on the
k
-thdesign variables.Now,bysettingalltheonstraintsandexeutingtheoptimization,thelargest
solution box is found for
f f d ≥ 95%
. Figure 6.7 shows the projetion of theomputed box for some design values based on the interative design spae
projetion and modiation. The volume of the omputed box is
7.5e −10 %
of the entire design spae volume. At the rst glane, it seems to be very
small, but it still gives us exibility in the proedure of the atuator design
andthe adjustmentoftheontrolsystemparameters. Inaddition, suhasmall
volumeprovesthatndingthesolutionboxmanuallyisalmostimpossible. The
intervals of the design variables (normalized), obtained from thesolution box,
are depited ingure6.8 .
Figure6.7: Projetion ofthesolution boxfor some onsidereddesign variables
Figure 6.8:ARS Solution intervals(normalized)
The parameters of the dynami feedforward ontrol and the feedbak
on-troller an be adjusted now with respet to the determined intervals. The
harateristi urveoftheARSstatifeedforwardontrol analsobelaying in
theyellow areaofgure 6.9.
Figure 6.9:Solution spae for the harateristi urve of the ARS stati
feed-forward ontrol
Computing requirements on veriation test variables of ARS
atuator
Aswasexplainedintheprevioushapter, thesolutionintervalsoftheatuator
design variables should beonverted into requirements on theveriation test
variablesoftheatuator. Theserequirementsanbethenspeiedinthe
prod-utrequirementdoument (PRD) oftheatuator,whihhasto besatisedby
thesupplier. Asmentionedbefore,theserequirementsareformulatedregarding
theveriation tests. The veriation testshave to be arriedout on the
test-rig depited ingure 6.10 . This has to be done for a seleted maximum rear
steering angle from its solution intervals. The following requirements will be
alulated for
δ r,max = 2.7 ◦
,themaximumoftherear steering angleinterval.Figure6.10: Test-rig setupof ARS
Requirements of the ARS atuator with respet to performane test
It is denitely possible to ompute solution intervals for the veriation test
variablesfromthesolutionintervalsofthedesignvariables aswell. However, it
is enough to dene the worst-ase of the atuator response to the veriation
tests with respet to the alulated solution intervals. Before alulating the
worst-aseofthe atuator fore-veloityharateristi urve, ithastobe
men-tioned thattheupperboundofthisdesignvariable(
F max
,v noload
)wasxedtoa onstant value during the optimization proess. This onstant value should
also begiven tosuppliers,astheymaynot onstrutanatuator withthe
per-formanemorethanthisonstantvalue. Jumpingbaktothealulationofthe
requirements on the veriation test variables, the worst-ase of the atuator
fore-veloity harateristi urve ours by the lower bounds of the solution
intervals of the
F max
,v noload
andv (F max )
. Figure 6.11 depits the minimumARS atuator fore-speed harateristi urve. The redarea is theprohibited
area. It means, a supplier must onstrut an atuator whose fore-veloity
harateristi urve isloated above this area.
Figure6.11: MinimumARS atuator fore-veloityharateristi urve
Requirements of the ARS atuator with respet to step response test
Generally,itisdesirabletohavelessovershoot,lessrisingtime,lessstabilization
time, andlesstime delay(quik responseto the stepinput). Thefatis, every
mehanial omponent, e.g. atuator, denitely has some overshoot, rising
time, stabilization time and time delay beause of it's physial onstrution
limits. Asa onsequene, the worst-ase of theatuator response to thestep
input with respet to the alulated solution intervals leads to the maximum
overshoot,maximumtimedelay,maximumstabilizationtime,maximumrising
time,andmaximumstationaryerror. Alltheseworst-asessubsequentlysatisfy
theupperand lowerboundsofthe design variables solutionintervals.
Insertingthe lowerbound ofthedamping ratiointerval
D
into equation5.15leads tothemaximumovershoot. Inthismanner,itguaranteesthesatisfation
D
Thestabilization time isdependent on the damping ratio andthetime
on-stant regardingequation5.16 . Aordingly,themaximumstabilization timeis
alulated fromthelowerbound ofthe determined interval ofthedamping
ra-tio,
D
,andthe upperboundofthetime onstant,T w
,solutioninterval;hene,themaximumstabilizationtime guarantees therealizationoftheupperbound
of the time onstant solutioninterval.
With respet to equation 5.17 , the worst-ase of the rising time ours by
the upper bounds of the damping ratio and the time onstant interval.
Con-sequently, the maximum rising time realizes the upper bound of the solution
intervals of
D
andT w
.Moreover, it is desirable to have a stationary error as small as possible. As
a result, the maximum possible permitted stationary error hasto be given to
suppliers. Thisarisesout oftheminimumbetween theratios oftheupperand
lowerbound of thesolutionintervalofthe stationarygain
K p
and1.Finally, the maximum time delay is measured by the upper bound of the
determined intervals of
T d
. Here, it should again be mentioned that thelowerbound ofthe
T d
wasxedduring theoptimization.It should also be notied that alulating all the worst-ase requirements
has been arried out regarding worst-ase of the fore-veloity harateristi
urve. In other words, after inserting all these upper and lower bounds of the
determined intervalsinto thetransferfuntionoftheARSatuator andsetting
thestepasinput,theworst-aseofthefore-veloityshouldbebuiltasthe
rate-limiteraftertheoutputofthetransferfuntion,showningure4.17 . Then,all
the requirementsareformulated for theoutput asfollows:
Table 6.4: ARSatuator step responserequirements
Fore speed urve parameters Value
Max. Time delay
25ms
Max. Overshoot
0.0%
Max. Stabilization Time
405ms
Max. RisingTime
319ms
Max. Stationary Error
0.3%
Requirements of the ARS atuator with respet to sine test
Here, the test is dened with an inreasing frequeny from 0.1 to 5 Hz. The
amplitude is onsidered as
1 ◦
. As mentioned in the previous hapter, there-quirements shouldbeformulatedon thebodediagram. It isdesirable thatthe
deayofthemagnitudeandthephaseshifttakesplaeinhigherfrequeny,but
not inanyhighfrequeny. Inthis manner,the atuator performs wellin lower
frequenies and the output of the atuator has no phase shift and amplitude
deay in low frequenies. As a onsequene, therequirements of this test are
dened for the upper and lower bound of the bode diagram. The better the
phase response, the better the performane of the atuator in the frequeny
domain is. Therefore, the phase response of the atuator may not be worse
than the determined worst-ase. Itmeans, the frequeny responseof theARS
atuator shouldbe equal or better than this worst response. The deayof the
magnitude and the phaseshift of the bode diagram ofa seond-order transfer
funtion ours at the natural frequeny of the system [45 ℄. The natural
fre-queny isequal to
ω 0 = T 1
w
. Inother words, the smallerthenatural frequeny
(the largerthe timeonstant)is, thesoonerthe dereaseof theamplitudeand
phase shift takes plae. Asa onsequene,the worst-aseof thebode-diagram
takesplae on the upperbound ofthe optimized timeonstant interval.
Con-erning the damping ratio
D
, the larger the damping ratio is, the earlier themagnitude response dereases. Subsequently, the worst-ase of the bode
dia-gramoursbytheupperboundoftheoptimizeddampingratiointerval;hene,
theworst-aseof thebode-diagram realizes the upperboundsofthe
D
andT w
solution intervals. However, as mentioned before, thedeay ofmagnitude and
phase shift may not our at ordinary high frequeny. So, the lower bounds
of
D
andT w
resultinthe upperbound ofthe bode-diagram. Aordingly, the permitted bode-diagram isbetween the two redareas depited gure6.12 .Figure6.12: Frequeny response requirement for theARS atuator